Support for the memory standard seems certain, but conflicting news has appeared about other specifications of the memory subsystem.
It’s clear that everyone wants to know what Nvidia’s next flagship model will be capable of, so rumor mills and individuals with industry resources on X will be sharing their thoughts with the general public as soon as possible. Due to the almost excessive power of the RTX 4090, many people are curious about what the direct successor can do, but the official release is so far away that there is no really meaningful discussion beyond rumours.
The first important point is the GDDR7 specification list officially published by JEDEC, based on which the initial computing capacity starts at 32 Gb/second and can peak at 48 Gb/second in the case of special accelerators. At the moment, this value is 23-24 on the current VGA generations, taking this as a basis, considering the VRAM, the progress is significant, but of course many other parameters can influence it. The design of the memory bus is one of them, and for now it is not clear in what structure the new GeForce generation may arrive.
The very first rabbit-tail post predicted a 512-bit memory bus for the top-of-the-line model, while lower-end versions could come with 256 or less VRAM wiring. The 384-bit design used on the RTX 4090 would therefore be omitted, which is a difficult prediction for the time being, since the 24 GB GDDR6X module integrated on the controller has a pleasant bandwidth; as far as playful use is concerned.
Even so, the printed circuit of the flagship is not easy (this also justifies the high price), to top it off with the 512-bit switch, while faster modules can be used; not very likely. At least not on the edition intended for the gaming market, the versions intended for workstations may have a larger memory bus, while gamers will get a simplified, 384-bit top model.
Thanks as always. GB203 is 256-bit and GB205 is 192-bit.
— kopite7kimi (@kopite7kimi) March 12, 2024
The 80 card coming under the 5090 model is expected to remain at 256 bits, like the RTX 4080, and the line will similarly continue for variants intended for lower price bands, as now. The RTX 5070 can manage 192 bits, and the mid-range GB206 and GB207 can use 128 bits.
In the absence of official confirmation, common sense and dreamlike data are mixed, it is worth starting from the RTX 40 data, adding the parameters coming from GDDR7. We don’t think there will be any big surprises, rather the increase in computing units will be significant on the GPU side. Much more mystery surrounds AMD’s RDNA4 plans, which may also include GDDR7, but the company is expected to interpret the word top model differently. If the rumors are true, then price-sensitive users can do well, but first let’s see the speed data, which may first be revealed during the summer Computex.
2024-03-13 16:03:38
#GeForce #RTX50 #series #ultrafast #GDDR7 #memory
